GTX 480 (825/1100 MHZ) vs. HD 5870 (975/1300 MHz)

[Ancient76]

http://alienbabeltech.com/main/?p=16997&page=22

This conclusion is a short one and it dictated itself. The HD 5870 overclocks solidly in many cases and once overclocked to about 975/1300 MHz it will in many cases catch the stock GTX 480 when performance is already reasonably close. However, the GTX 480 when it is overclocked also turns into a performance monster and in many cases now runs away from even the overclocked HD 5870. This bodes very well for GF100 Fermi architecture and indicates that NVIDIA has very solid scalable new architecture to build on although it is still on an improving process.

[CaptNKILL]

Yeah, I decided that I would buy a 480 if it could be overclocked without setting fire to my computer. The overclocked performance I've seen on various sites is extremely impressive.

The heat and power consumption is just unacceptable for me though.

[Logical]

Hmm intersting, theres a couple of those benchmarks were the stock GTX480 outperforms the OC'd HD5870.

I am still going to wait for the next revisions of the fermi before i purchase one.

[brunner]

Seems like there's a much bigger jump when the 480 is overclocked. It's been a mystery for awhile why there aren't much performance benefits even when the 5870 is overclocked quite a bit.

[MUYA]

Originally Posted by brunner

Seems like there's a much bigger jump when the 480 is overclocked. It's been a mystery for awhile why there aren't much performance benefits even when the 5870 is overclocked quite a bit.

Could be a bottleneck somewhere in the actual GPU/engine?

[Ninja Prime]

Originally Posted by MUYA

Could be a bottleneck somewhere in the actual GPU/engine?

Memory bandwidth would be my guess. 1200-->1300 at 256 bit isn't much bandwidth increase(+12.8GB/s). However, on the 480 they used here, 924 --->1100 at 384 bit is more than twice as much bandwidth gain, about 2.5 times or so(+33GB/s). Mainly because NVs GDDR5 memory controller sucks, they are actually downclocking the slowest GDDR5 to make it work, but it leaves them with more memory overclocking headroom.

[brunner]

Ninja, I think you're right. There's a significant jump when the 5970 is overclocked to 2 x 5870 levels because there's like a 200 mhz jump in memory. However, most 5870s are limited to 1300mhz memory. It's very likely that there's just no benefit to overclocking the core unless there is a bigger bump in memory speed.

[Viral]

Percentage wise, the core overclock on the 480 is 17.8% and on the 5870 only, 14.7%. Then as said there's the huge difference in memory bandwidth increase.

Still, there may be some sort of better scaling on the 480 with clock speed. Afterall, it is a new architecture while RV870 is the supposed last revision of R600.

[azanon]

Originally Posted by Ancient76

http://alienbabeltech.com/main/?p=16997&page=22

This conclusion is a short one and it dictated itself. The HD 5870 overclocks solidly in many cases and once overclocked to about 975/1300 MHz it will in many cases catch the stock GTX 480 when performance is already reasonably close. However, the GTX 480 when it is overclocked also turns into a performance monster and in many cases now runs away from even the overclocked HD 5870. This bodes very well for GF100 Fermi architecture and indicates that NVIDIA has very solid scalable new architecture to build on although it is still on an improving process.

Man, this card is about to overheat and set on fire at default. It's hard to imagine that someone would even consider overclocking an already overheating card with a fan that's despirately trying to cool it.

What I'm interested in is any link where someone tried to DOWNCLOCK the card to the speed it probably should be running at.

[Vardant]

Reviewers say one thing, but the actual owners are saying something else. It all boils down to reviews being done in an open enviroment. These cards behave differently, when put in a case. The temperatures are much lower and so is the noise. So far, people are reporting temperatures around 70°C or 80°C in heavy scenarios. That's quite a jump from the numbers we were told

[Ancient76]

Originally Posted by azanon

Man, this card is about to overheat and set on fire at default. It's hard to imagine that someone would even consider overclocking an already overheating card with a fan that's despirately trying to cool it.

What I'm interested in is any link where someone tried to DOWNCLOCK the card to the speed it probably should be running at.

?

Dude, it isn't anything like that at all. I should know, I actually have one. After an hour of Metro 2033 DX11 maxed out settings. Card is in the high 70's using EVGA Precision tool. Fan never exceeds 66% and is not audible at all.

The cards have had an updated bios from the initial cards sent out to reviewers. There has been a big improvement, that's for sure.

[azanon]

Originally Posted by Vardant

Reviewers say one thing, but the actual owners are saying something else. It all boils down to reviews being done in an open enviroment. These cards behave differently, when put in a case. The temperatures are much lower and so is the noise. So far, people are reporting temperatures around 70°C or 80°C in heavy scenarios. That's quite a jump from the numbers we were told

I don't leave my case open either, but I'm personally going on relative comparisons. By that I mean, I personally own a GTX 285, and I briefly owned a GTX 295. The 295 is about as loud as I'd want to have to listen to with a case closed.

Assuming the GTX 480 is louder than a 295 - a solid assumption based on the reviews I've read - then its just too loud IMO.

....

I've seen the 94-95C repeated and reported by several long-standing review sites (think the top names here) who probably have near perfected their measuring techniques. Who am I going to trust?